Leadership and incentives∗
Alexander W. Cappelen
Erik Ø. Sørensen
Bjørn-Atle Reme
Bertil Tungodden
April 7, 2015
Abstract
We study how leader compensation affects public goods provision. We
report from a lab experiment with four treatments, where the base treatment was a standard public goods game with simultaneous contribution decisions, while the three other treatments allowed participants to volunteer to
be the leader in their group and make their contribution before the others.
In the three leader treatments, we manipulated the level of compensation
given to the leader. Our main finding is that a moderate compensation to
the leader is beneficial, it increases the average contribution relative to both
a situation where the leader is not compensated and a situation without a
leader. A further increase in the leader compensation, however, is detrimental to public goods provision; it attracts more free-riders and creates a social
crowding- out effect. Finally, we report from a survey showing that the social crowding- out effect is also present in the population at large. We argue
that the main findings of the paper are important in many real life settings
where we would like to use economic incentives to encourage people to lead
by example.
∗ We
have received extremely valuable comments and suggestions from three anonymous referees, Thomas Dohmen, Tore Ellingsen, Urs Fischbacher, Uri Gneezy, Hans K. Hvide, Sebastian
Kube, Erling Moxnes, Andreas Nicklisch, Øivind Schøyen, Matthias Sutter, Jean-Robert Tyran,
Eline van der Heijden, and Marie Claire Villeval. We would also like to thank Eli Birgitte S. Bergsmark, Ida Elise S. Bergsmark, Sebastian Fest, Solfrid Dahl Helgesen, Linn-Therese Holsen, John
McGregor, Felix Morency-Lavoie, Katrine Nødtvedt, Andreas Olden, Easa Sahabeh Tabrizi, Ana
B. Santos, Giancarlo Spagnolo, and Trond Halvorsen for research assistance and Bjrn Ivar Grøtt˚a,
˚
Sarah Marie Søiland, Katrine Asheim,
and Atle Askeland for IT-support and administrative assistance. The project was financed by support from the Research Council of Norway, research grant
185831, and the Center for the Study of Mind in Nature, University of Oslo, and administered by
The Choice Lab, NHH Norwegian School of Economics.
1
1
Introduction
The possibility to lead by example is potentially important in many public goods
settings. In the household, at the workplace, in the local community, in business
and politics, it is often possible for agents to take the lead and announce their
contribution to the public good before others do. Since many people are conditional contributors in public goods contexts (Fischbacher and G¨achter 2010),
first movers can potentially inspire others to contribute. But how can people be
motivated to take the lead and when is leading by example likely to be effective?
Evidence from economic experiments with the public goods game suggests
that leading by example has the potential to increase contributions of others, particularly when leadership is voluntary (Moxnes and van der Heijden 2003, Arbak
and Villeval 2013, G¨achter et al. 2010b, Haigner and Wakolbinger 2010, Rivas
and Sutter 2011). The same studies, however, also find that leaders typically end
up worse off than their followers, since followers tend to contribute less than the
leader. Thus, in contexts where it is a voluntary decision to become a leader, one
may face the challenge that no one will step up and show leadership.
This problem is familiar in many real life settings. At the work place, people
may hesitate to take the leading role in organizing valuable social activities, fearing that their co-workers will not contribute to the same extent. Similarly, parents
may avoid taking the lead in organizing a school activity for the children, worrying that the other parents will not follow up. More generally, whenever there is
the possibility of taking the lead in a public goods setting, people may hesitate to
be the first mover.
The classical solution suggested by economists to the problem of undersupply
is to strengthen economic incentives. By now it is well recognised, however, that
the introduction of economic incentives may undermine the intrinsic motivation
of individuals, and that the net effect may be a reduction of the overall motivation (Frey and Oberholzer-Gee 1997, Gneezy and Rustichini 2000, Fehr and Falk
2002, B´enabou and Tirole 2003, Gneezy et al. 2011). Thus, it is not obvious that
offering leader compensation would strengthen people’s willingness to volunteer
as leaders; the economic incentive may crowd out any intrinsic motivation to step
up and show leadership. In the public goods context, however, there is a further
concern regarding the introduction of economic incentives for the leader; it may
crowd out the intrinsic motivation of the other members of the group. We coin
the effect that an increase in compensation to the leader reduces the contribution
of followers, as the social crowding-out effect. This effect may work through
different channels; leader compensation may make the others less motivated to
2
contribute because they care about inequality in the group or because they envy
the leader or feel disappointment at not themselves being the leader.
To study the effect of leader compensation we designed a lab experiment with
a sequence of one-shot public goods games with random rematching. In addition
to a treatment without a leader, we had three treatments where the participants
could volunteer to be the leader whose contribution would be announced before
the other members of the group decided on their contribution. The three treatments differed only in the compensation given to the leader (zero, moderate and
high). This design allows us to study whether leader compensation represents a
promising approach to public goods provision in settings where it is possible for
an agent to move before others, and allows us to shed light on whether leader
compensation creates a social crowding-out effect. Further, to study the generality of the social crowding-out effect, we also conducted a large-scale survey on a
representative population. In the survey, we asked the participants to make a hypothetical choice of how much to contribute to a public good in a situation where
someone else had been selected to be the leader of the group. Also in the survey,
we introduced three treatments that differed only in the compensation paid to the
leader (zero, moderate, and high).
The paper offers four main findings. First, we show that leader compensation does indeed improve public goods provision. Second, we find that the optimal strategy is to pay leaders a moderate compensation. In the lab experiment,
moderate leader compensation increases average contribution both compared to
the base treatment without a leader and to the treatment where the leader is not
compensated. A further increase in the leader compensation, on the other hand,
causes a considerable drop in average contributions. Third, we show that a high
leader compensation is detrimental because it attracts more free-riders and creates
a social crowding-out effect by reducing the followers’ willingness to contribute.
Finally, we show by a large-scale survey that the social crowding-out effect is
present in the population at large; people are less willing to contribute to a public
good in a hypothetical choice situation when they are told that the leader is highly
compensated. In sum, the paper shows that leader compensation may be a viable
strategy for increasing public goods provision, but also that it may initiate selection and social crowding-out effects that make it optimal to keep the compensation
at a moderate level.
Our paper contributes to a growing experimental literature on voluntary leadership in public goods games. The first paper to study voluntary leadership experimentally is Arbak and Villeval (2007), now published as Arbak and Villeval
(2013). They examine what drives people to lead, and find that strategic and altru3
istic concerns are important. In line with their findings, we show that mostly otherregarding participants volunteer to be the leader when there is no leader compensation, while free-riders are particularly attracted to be the leader when there is
high leader compensation. Our paper also relates to G¨achter et al. (2010a), who
show that groups perform best when they are led by those who are cooperatively
inclined. Their finding highlights the importance of understanding how leader
compensation affects selection into the leader role. Finally, our paper is inspired
by Haigner and Wakolbinger (2010) and Rivas and Sutter (2011), who establish,
using partner designs, that voluntary leadership is better than imposed leadership,
but also that most participants do not want to lead when there is no leader compensation. In our experiment, we too observe that few participants volunteer when
there is no leader compensation and as a result the groups do not perform better in
this treatment than in the baseline treatment without a leader. The main contribution of the present paper is therefore to study whether leader compensation can be
used to secure voluntary leadership and increase public goods provision, and to
investigate the potential mechanisms through which leader compensation affects
the behavior of both leaders and followers.1
The structure of the paper is as follows: Section 2 presents the design of the lab
experiment, Section 3 reports treatment effects, Section 4 investigates potential
mechanisms through which leader compensation affects cooperation, Section 5
reports from the survey, and Section 6 provides some concluding remarks.
2
The lab experiment
The lab experiment consisted of four treatments of a repeated public goods game
with random rematching. The base treatment was a standard public goods game
with simultaneous contribution decisions, while the three other treatments allowed
participants to volunteer to be an “early contributor” in their group. Being an early
contributor meant that one’s contribution would be announced to the other group
members before they made their contribution decision. We shall in the following
refer to the “early contributor” as the “leader,” though the word “leader” was never
used in the experiment.
1 Leadership
has many important aspects that are not investigated in the present paper, including granting the leader the authority to reward or punish (G¨urerk et al. 2009, G¨uth et al. 2007),
asymmetric payoff structures (G¨achter et al. 2010a, Gl¨ockner et al. 2011, Levati et al. 2007, van der
Heijden and Moxnes 2013) and asymmetric information (Potters et al. 2007).
4
2.1
The participants and procedures
We recruited participants among students at the NHH Norwegian School of Economics. A total of 508 subjects participated in 13 sessions.2 Each session lasted
approximately 90 minutes and the average payment was 365 NOK (about 60
USD), plus a show-up fee of 100 NOK. The experiment was conducted in a computer lab using a web-based interface and was double blind.
Upon arrival the participants were informed about the rules of conduct, given
an overview of how the experiment would proceed, and introduced to the general
public goods game. The participants were also informed that they would remain
anonymous throughout the experiment. A copy of the instructions was available
on the desk of each participant. After the introduction, the participants were given
a set of control questions in order to ensure that they had understood the nature of
the public goods problem. All ten control questions had to be answered correctly
before a participant could proceed to the actual experiment. After the experiment
was conducted, the participants were asked some questions about what motivated
their contribution decision and whether they felt envious or disappointed if they
volunteered, but were not selected, to be the leader.3
All payments were made in cash immediately after the experiment. Special
care was taken so that the payment procedure ensured anonymity. The computer
assigned a payment code to each of the participants, and a group of assistants who
were not present in the lab during the experiment prepared envelopes containing
the payments corresponding to each payment code. After bringing the envelopes
to the lab the assistants left immediately, and the envelopes were handed out in
accordance with the payment codes. This procedure was explained to all participants at the start of the experiment.
2 We
conducted ten sessions with 272 participants in 2011 and three sessions with 236 participants in 2013. In 2011, we randomly allocated treatments into sessions with the restriction that no
treatment was implemented only in the morning or only in the afternoon; all participants in a given
session thus participated in the same treatment. In 2013, we ran larger sessions and the participants
in each session were randomly allocated to one of the four treatments. We thus have 13 matching
groups in 2011 and 12 matching groups in 2013. Details about the number of participants in each
session and treatment are given in Table A1 in Appendix 1. The instructions and screenshots are
provided in Appendix 2.
3 These questions were only asked in the 2013 sessions, inspired by comments from a referee.
5
2.2
The treatments
All four treatments consisted of ten rounds of a public goods game with random
rematching of groups after each round. The participants were randomly assigned
to groups of four and in each round they each received 20 points (1 point = 1
NOK) that they could either keep or invest in a public good. Each participant’s
return from the public good was 0.4 times the group’s total investment in the
public good. After each round the participants were informed about their payoff
in this round and about their group’s total contribution to the public good.
Our base treatment (B-treatment) was a standard public goods game where
all group members simultaneously chose their contribution to the public good.
This treatment was a replication of the experiment presented in Fischbacher and
G¨achter (2010). In the three leadership treatments the participants could volunteer to be the leader. If more than one group member volunteered, the leader
was randomly chosen among the volunteers and asked to state his contribution.
The leader contribution was announced to the three other group members before
they decided on their contribution.4 The participants were not informed about
the number of volunteers. If no one volunteered to be the leader, the round was
played with simultaneous contribution decisions. The only difference between
the three leadership treatments was the level of compensation given to the leader.
In the “no compensation” treatment (NC-treatment) the leader received no compensation.5 In the “medium compensation” treatment (M-treatment) and the “high
compensation” treatment (H-treatment), the leader received four and twelve points
respectively to his private account as compensation for being the leader. Importantly, the leaders in the M-treatment and the H-treatment could not use the leader
4 In
the 2011 sessions, we imposed time limits to avoid unnecessary delays. The participants
were given 60 seconds to decide whether they wanted to be leaders and 90 seconds to decide how
much to contribute. Ten points were subtracted if they did not decide within the time limit. This
was made clear to the participants before starting and the time spent was visible on the screen
during the experiment. The penalty was only effected in 1.2% of the situations (33 times). One
referee was concerned, however, that the time limit might give the impression that the participants
had to rush to get the high payoff in the high compensation treatment, without thinking about
what leading a group means. To accommodate this concern, there were no time limits in the 2013
sessions.
5 The NC-treatment is related to the design in Rivas and Sutter (2011), who also study voluntary leadership without compensation. There are two important differences between their study
and our NC-treatment; first, they consider a repeated game with partner matching, and second they
let subjects go ahead as leaders if they entered a contribution more quickly than others. These differences may explain why they find a stronger effect of voluntary leadership without compensation
than what we observe in the present study.
6
compensation to contribute to the public good. Thus, the payoff structure in the
public goods game itself was identical in the four treatments.
To summarize, a participant’s payoff in each round can be represented in the
following way,
4
πi = 20 − ci + 0.4 ∑ c j + di · ft ,
(1)
j=1
where ci ∈ [0, 20] is participant i’s contribution to the public good, di is a binary
variable indicating whether the participant was a leader, and ft indicates the compensation paid to the leader (which depended on the treatment).6
2.3
A conditional contribution experiment
All participants took part in a second experiment similar to the P-experiment in
Fischbacher and G¨achter (2010).7 In this P-experiment, the participants first decided on how much they wanted to contribute in a one-shot public goods game.
The public goods game was the same as in the base treatment (except that in this
case 1 point = 5 NOK). After deciding on the contribution in the one-shot game, all
participants were asked to fill out a conditional contribution table. For each possible average contribution level of the other three participants (rounded to whole
integers), the participant was asked to specify how much he would want to contribute to the public good. When all participants had made their decisions, three
of the four participants in a group (randomly selected) contributed according to
their decision in the one-shot unconditional game. The contribution of the fourth
6 After
making their contribution decision participants were also asked what they believed the
other participants would do. Leaders were asked what they believed the average contribution of
the followers would be, given their own contribution. Followers were asked what they believed
the average contribution of the two other followers would be, given the leader’s contribution.
Participants in situations without a leader were asked what they believed the average contribution
of the three other participants would be. The questions were incentivized and the participants
received three points for each correct answer, two points if their answer deviated one point from
the correct answer, and one point if the answer deviated two points. The beliefs data confirm
almost exactly the analysis in Fischbacher and G¨achter (2010), where it is shown that declining
cooperation in a public goods game is driven by people being imperfect conditional cooperators.
We also find that the beliefs of followers about others’ contribution are strongly correlated with the
leader’s contribution. It is, however, not possible to identify whether the underlying mechanism
in this case is that the follower is inspired by the leader’s contribution and is therefore prone to a
false consensus effect, or that the follower is not himself inspired but believes that others are so.
7 The P-experiment was conducted after the main experiment in 2011, but, on the suggestion of
a referee, before the main experiment in 2013.
7
participant was determined by his conditional contribution table and the average
of the three other participants.
The P-experiment provides us with an independent measure of the social preferences of the participants. In the following, we classify a participant as a freerider if he did not want to contribute for any level of the average contribution of
the others in the group, otherwise he is classified as an other-regarding participant.
3
Results
The left panel in Figure 1 shows how the contributions to the public good evolved
in the experiment. We observe that in all treatments, there is a declining trend in
contributions in later rounds. In the B-treatment, the average contribution is 31.7%
of the endowment in the first round but falls to 9.9% of the endowment in the last
round, which is very similar to the pattern observed in Fischbacher and G¨achter
(2010).8 The introduction of the possibility to volunteer to be the leader without any compensation (NC-treatment) slightly increases average contributions in
most rounds compared to the B-treatment. Offering a leader compensation, however, significantly increases average contributions relative to the B-treatment in all
rounds (with the exception of the last round for the H-compensation treatment).
A moderate leader compensation is particularly productive; the average contribution in the M-treatment is above the average contribution in the H-treatment in all
rounds (with the exception of the ninth round).
[ Figure 1 about here]
The right panel in Figure 1 presents treatment effects on the average contribution to the public good across rounds.9 We observe that the introduction of a
8 Fischbacher
and G¨achter (2010) show that the declining trend of contributions reflect that
people are “imperfect conditional cooperators” who match other’s contributions only partly, rather
than learning. This is consistent with the well-known restart effect in public goods games, where
contribution levels increase when the game is restarted (Andreoni 1988). The fact that we had
an extensive protocol with control questions also makes it unlikely that learning is important in
our experiment, and thus we do not focus particularly on the behavior in the final rounds in the
analysis.
9 From Table A2 in Appendix 1, we observe that the treatment results hold if we consider
rounds 1-5 and 6-10 separately, even though they are significantly weaker in the second part of the
experiment. Figure A1 in Appendix 1 furthermore shows that we have the same treatment results
in the first round, before any interaction has taken place. Finally, Figure A2 and Table A3 in
8
moderate compensation to the leader increases contributions (p < 0.01); the average contribution in the M-treatment is 63% higher than in the NC-treatment and
90% higher than in the B-treatment.10 A further increase in leader compensation,
however, is detrimental to the public good provision; the average contribution in
the H-treatment is significantly lower than in the M-treatment (p < 0.01), but still
higher than in the NC-treatment (p = 0.012). Finally, we observe that the average
contribution in the NC-treatment is slightly above the average contribution in the
B-treatment, but this difference is not statistically significant (p = 0.30).
In sum, this analysis provides three important insights. First, the introduction
of leader compensation increases contributions to the public good; second, a too
high leader compensation can be counterproductive; third, the reliance on participants volunteering to be leaders without receiving any compensation is not a
viable strategy for ensuring public goods provision.
4
Mechanisms
In this section, we investigate potential mechanisms through which leader compensation affects cooperation. We first look at how the level of leader compensation affected the recruitment of leaders, before we discuss how it affected leader
and follower behavior.
4.1
Recruitment of leaders
As shown in the left panel in Figure 2, the level of compensation had a significant positive effect on the participants’ willingness to be leaders.11 In the NCtreatment, participants volunteered to be a leader in only 23% of the situations,
whereas they did so in 63.4% and 94.4% of the situations in the M-treatment
Appendix 1 consider the treatment effects separately for the sessions conducted in 2011 and 2013,
respectively. We observe that the pattern is strikingly similar, even though there is a statistically
significant difference in the level of contributions between 2011 and 2013. See also Tables A4-A5
in Appendix 1 for more detailed descriptive statistics.
10 If contribution is measured relative to endowment, the differences in contribution are 12 percentage points (M-treatment versus NC-treatment) and 14 percentage points (M-treatment versus
B-treatment); in terms of average earnings, the differences are about 6% (M-treatment versus NCtreatment) and 8% (M-treatment versus B-treatment). Throughout the paper, reported p-values are
from t-tests of equality, where standard errors have been corrected for clustering on individuals
where appropriate, using the sandwich estimator of Liang and Zeger (1986).
11 In Figure A3 in Appendix 1, we provide disaggregated data by round.
9
and the H-treatment, respectively. As shown in the right panel, this implied that
there was almost always a leader in the group in the treatments with leader compensation (97.1% of the groups in M-treatment, 99.7% of the groups in the Htreatment). In contrast, we only had leaders in 64.4% of the groups in the NCtreatment.12
[ Figure 2 about here]
The introduction of leader compensation thus largely solved the problem observed in many of the groups in the NC-treatment, that no one stepped up and
made the first move to contribute to the public good.
4.2
Leader behavior
We now turn to a discussion of how leader compensation affected leader behavior.
As shown in Figure 3, the average leader contribution in the H-treatment was
significantly lower than in the M-treatment (p = 0.04), whereas we do not find a
statistically significant difference between the M-treatment and the NC-treatment
(p = 0.70) or the H-treatment and the NC treatment (p = 0.17).
[ Figure 3 about here]
Leader compensation may affect both the selection of leaders and the motivation of those who become leaders.13 A high compensation may attract more freeriders (as measured by the P-experiment) to volunteer, as indeed we observe in
Figure 4.14 In the NC-treatment, a very small share of the free-riders volunteered
to be leader (9.7%), whereas almost all of them did so in the H-treatment (92.6%).
Thus, the introduction of leader compensation changed the composition of leaders, with a larger share of free-riders in the M-treatment and the H-treatment.15
This may partly explain the lower level of leader contribution observed in the
H-treatment.
12 In all pairwise comparisons of the treatments, the difference in the share of participants that
volunteered to be leaders and the difference in the share of groups with a leader are statistically
significant (p < 0.01).
13 See also G¨
achter et al. (2012).
14 We observe the same pattern if we consider the share of free-riders that volunteered to be
leaders on the share of other-regarding participants, see Figure A4 in Appendix 1. In total, 21.3%
of participants are classified as free-riders.
15 The selection into the leader role is also affected by the participants’ beliefs about the likelihood of others volunteering to become leader, how other’s would behave as leaders, and how
followers’ would respond to the leader’s contribution.
10
[ Figure 4 about here]
Leader compensation may, however, also affect the motivation of any given
leader. If a leader believes that the leader compensation makes it harder to lead
by example, then he may become less motivated to contribute. At the same time,
leader compensation may also introduce a reciprocal motive, where morally motivated leaders may feel a stronger moral obligation to contribute to the public
good. The net effect of leader compensation on leader contribution is therefore
hard to predict.
Overall, as shown in Figure 3, the net effect of leader compensation on leader
contribution is slightly positive for the M-treatment compared to the NC-treatment,
but negative for the H-treatment.
4.3
Follower behavior
How did the followers respond to leader behavior? Figure 5 reports the correlation between the leader’s contribution and the followers’ contribution, where
we observe a positive relationship in all treatments; a higher contribution by the
leader is, on average, followed by a higher contribution by the followers. We also
observe that it is very rare that the followers contribute more than the leader, this
happened only in 6.1% of the situations. The leader’s contribution thus appears
to have served as an upper ceiling for the followers, and on average the followers’
contribution was 42.6% of the leader’s contribution (see Table A4 in Appendix 1).
[ Figure 5 about here]
A social crowding-out effect from leader compensation on follower behavior may affect both the margin and the level of follower contribution. In Table
A6 in Appendix 1, we show that there are no statistically significant differences
across treatments in how followers respond to a marginal increase in leader contribution (p = 0.20). Further, combining the marginal response of followers with
observed treatment differences in leader contributions, cannot fully account for
the observed treatment differences in follower behavior. In particular, leaders in
the M-treatment contribute with 1.67 points more than leaders in the H-treatment,
which, using the estimated marginal response of followers (column 2, Table A6),
translates into a 0.55 points difference in follower contributions. The actual difference in followers’ contributions between the M-treatment and the H-treatment,
however, is 1.29 points, which means that the marginal response can only account
11
for about 42% of the social crowding-out effect from high leader compensation.
This suggests that there is also a social crowding-out level effect on followers’
contribution and, consistent with this, we observe in columns 2 and 3 in Table A6
that there are statistically significant treatment differences in the estimated levels
of followers’ contribution (p < 0.01 in both cases).
The observed treatment differences in followers’ contribution cannot be given
a causal interpretation, since the participants self-select into the role of followers and leaders in treatments. To shed more light on how leader compensation
affected followers in the M-treatment and the H-treatment, we conducted a short
questionnaire at the end of the experiment. In the questionnaire, we asked them
whether the payment to the leader affected their willingness to contribute to the
public good as a follower, whether they were concerned about inequality in the
group when making a contribution, whether they felt envy towards the leader, and
whether they were disappointed not themselves to be the leader. Since all participants were follower in at least one group in the M-treatment and the H-treatment,
there are no composition effects when comparing these responses for the two treatments. Thus, average differences in responses cleanly identify the causal effect of
leader compensation on follower motivation.
Figure 6 reports the responses in the questionnaire, where we observe that
there are systematic differences between the M-treatment and the H-treatment.
In particular, we find that the leader compensation in the H-treatment made the
followers significantly less motivated to contribute than in the M-treatment (p =
0.04). This is consistent with the level effect observed above, namely that the
followers’ contribution as share of leader contribution is significantly lower in
the H-treatment than in the M-treatment, and thus provides further evidence of a
social crowding-out effect from high leader compensation. This social crowdingout effect appears to be partly driven by envy (p < 0.01) and disappointment (p <
0.01) being stronger in the H-treatment than in the M-treatment, whereas we do
not see any treatment difference in the concern for inequality (p = 0.72).
[ Figure 6 about here]
The questionnaire responses thus suggest that there is a social crowding-out
effect of high leader compensation in the sense that a high compensation makes
the followers less motivated to contribute to the public good. The responses are
furthermore consistent with the reduced willingness to contribute being a result of
envy and disappointment among those who did not become a leader. We now turn
to a discussion of whether this effect is also present in the population at large.
12
5
Survey evidence
To shed further light on the effect of leader compensation on follower behavior,
we conducted a survey in collaboration with Norstat, one of the leading datacollection agencies in Europe. We surveyed 1952 respondents from Norstat’s
subject pool, where the respondents were chosen such that they would constitute
a representative sample of the Norwegian adult population (18 years and older).16
The participants were presented with the following hypothetical choice situation:
“You live in an apartment building with 20 apartments where the
residents volunteer for maintenance work six Saturdays each year.
The work done is painting, gardening and similar tasks. Each volunteer day lasts 5-6 hours. Several residents, you included, volunteered
to lead the planning and execution of the initiative. Using a simple
lottery, another resident ended up with the leadership responsibility.
The leader will receive no/moderate/high monetary compensation for
doing the job, while the other residents taking part are not compensated.
Assuming that you are not prevented from participating on any of
the six days that are scheduled for volunteer work in your building,
on how many of the days do you think that you would participate?”17
The participants were randomly allocated into one of three treatments, where
the only difference between the treatments was the announced amount of leader
compensation (no/moderate/high). Thus, in line with the lab experiment, the survey design captured a public goods situation where a voluntary leader is given
different levels of compensation. An advantage of the survey design is that it ensures that all participants face a situation with a leader, including the participants
in the no compensation treatment. Furthermore, since we state that the leader
participates all six days, we also ensure that the leader contribution is the same
16 Table
A7 in Appendix 1 provides sample characteristics.
survey was conducted in two rounds, where we used slightly different formulations to
test for robustness. In particular, we varied how we stated the monetary compensation and the
choice set. In one round, we used “no/a small/a very high” monetary compensation and a choice
set of zero to six days, in the other round we used actual sums in NOK when stating the monetary
compensation and a choice set of zero to four days. The responses were strikingly similar in the
two rounds and we thus only report the aggregate results. For simplicity, we here refer to the
treatment variation in the same way as in the lab experiments (no/moderate/high compensation).
17 The
13
across treatments. The survey design can therefore cleanly identify the level effect of introducing leader compensation on the followers’ self-reported motivation
for contributing.18
[ Figure 7 about here]
Figure 7 shows average reported participation in voluntary work by treatment.
Consistent with the lab findings, we observe that a high leader compensation
causes a significant drop in the willingness to contribute to the public good, both in
a comparison with the no compensation treatment (p < 0.01) and in comparision
with the moderate compensation treatment (p < 0.01). The survey thus provides
clear evidence of a social crowding-out effect in the level of public good contribution in the population at large. A moderate leader compensation also causes a
decrease in public goods provision, but this effect is not statistically significant
(p = 0.27).
6
Concluding remarks
We have shown that monetary compensation to leaders can be used to increase
public goods provision, but also that it may create a social crowding-out effect of
the willingness to contribute to the public goods. We believe that these insights
may be of great importance in a wide range of real life situations.
It has been estimated that nearly a billion people are involved in voluntary
work throughout the world, and the voluntary sector has increasingly been recognised to play a fundamental role in society, and particularly in the provision of
public goods (Salamon et al. 2011). A crucial question in the organization of
voluntary work is how to motivate people to take the lead and inspire others to
follow. An example from daily life is the question of how to organize a youth soccer team, which typically requires input from many parents. How can we make
sure that someone is willing to take the responsibility of being the first mover
in getting the team established? The present paper has investigated one possible
approach, namely to provide a monetary compensation to the leader of the team.
Our experiment suggests that such a compensation may be beneficial for the effort
put into the organization of the soccer team. It may make it more likely that someone steps up, and the presence of a leader will most likely generate more effort
18 Since
the design does not vary the leader contribution, the survey cannot shed light on how
leader compensation affects the followers’ responsiveness to the leader’s behavior (the marginal
effect).
14
from the other parents. But this strategy must be implemented with care. A too
high compensation to the leader of the team may backfire, both by weakening the
power of the example and by attracting the wrong leaders.
Similar examples abound in the public and the private sector. For example,
how should we think of monetary compensation to the department head at a university? A high level of compensation will probably attract many candidates to the
job (including candidates with the wrong qualities), but it may also undermine the
willingness of the rest of the department members to contribute to the provision of
public goods in the department. The fact that the head of department works very
hard to create a well-functioning department may not inspire others to make the
same effort if they know that he is highly compensated for taking the lead.
We find that compensating leaders involved in public goods provision may
have both positive and negative effects, and the relative importance of these effects
may depend on the level of compensation offered to the leader. In particular,
we have shown that a high compensation to leaders may generate a significant
social crowding-out effect of moral motivation that may be detrimental for public
goods provision. Further research, however, is needed to understand how these
effects work in different types of settings, where also other aspects of leadership
are potentially involved.
References
Andreoni, James (1988). “Why free ride? Strategies and learning in public goods experiments,” Journal of Public Economics, 37(3): 291–304.
Arbak, Emrah and Marie-Claire Villeval (2007). “Endogenous leadership: Selection and
influence,” GATE Working Paper 2007-07 and IZA Discussion Paper 2732, Bonn.
Arbak, Emrah and Marie-Claire Villeval (2013). “Voluntary leadership: motivation and
influence,” Social Choice and Welfare, 40(3): 635–662.
B´enabou, Roland and Jean Tirole (2003). “Intrinsic and extrinsic motivation,” Review of
Economic Studies, 70(1): 489–520.
Fehr, Ernst and Armin Falk (2002). “Psychological foundations of incentives,” European
Economic Review, 46(4-5): 687–724.
Fischbacher, Urs and Simon G¨achter (2010). “Social preferences, beliefs, and the dynamics of free riding in public goods experiments,” American Economic Review, 100(1):
541–556.
Frey, Bruno S. and Felix Oberholzer-Gee (1997). “The cost of price incentives: An em-
15
pirical analysis of motivation crowding-out,” American Economic Review, 87(4):
746–755.
G¨achter, Simon, Benedikt Herrmann, and Christian Thoeni (2010a). “Culture and cooperation,” Philosophical Transactions of the Royal Society B: Biological Sciences,
365(1553): 2651–2661.
G¨achter, Simon, Daniele Nosenzo, Elke Renner, and Martin Sefton (2010b). “Sequential
vs. simultaneous contributions to public goods: Experimental evidence,” Journal of
Public Economics, 94(7-8): 515–522.
G¨achter, Simon, Daniele Nosenzo, Elke Renner, and Martin Sefton (2012). “Who makes
a good leader? cooperativeness and leading-by-example,” Economic Inquiry, 50(4):
953–967.
Gl¨ockner, Andreas, Bernd Irlenbusch, Sebastian Kube, Andreas Nicklisch, and HansTheo Normann (2011). “Leading with(out) sacrifice? A public-goods experiment
with a privileged player,” Economic Inquiry, 49(2): 591–597.
Gneezy, Uri, Stephan Meier, and Pedro Rey Biel (2011). “When and why incentives
(don’t) work to modify behavior,” Journal of Economic Perspectives, 25(4): 191–
210.
Gneezy, Uri and Aldo Rustichini (2000). “Pay enough or don’t pay at all,” Quarterly
Journal of Economics, 115(3): 791–810.
¨ ur, Bernd Irlenbusch, and Bettina Rockenbach (2009). “Motivating teamG¨urerk, Ozg¨
mates: The leader’s choice between positive and negative incentives,” Journal of
Economic Psychology, 30(4): 591–607.
G¨uth, Werner, M. Vittoria Levati, Matthias Sutter, and Eline van der Heijden (2007).
“Leading by example with and without exclusion power in voluntary contribution
experiments,” Journal of Public Economics, 91(5-6): 1023–1042.
Haigner, Stefan D. and Florian Wakolbinger (2010). “To lead or not to lead: Endogenous
sequencing in public goods games,” Economics Letters, 108(1): 93–95.
van der Heijden, Eline and Erling Moxnes (2013). “Leading by example to protect the
environment: Do the costs of leading matter?” Journal of Conflict Resolution, 57(2):
307–326.
Levati, M. Vittoria, Matthias Sutter, and Eline van der Heijden (2007). “Leading by example in a public goods experiment with heterogeneity and incomplete information,”
Journal of Conflict Resolution, 51(5): 793–818.
Liang, Kung-Yee and Scott L. Zeger (1986). “Longitudinal data analysis using generalized
linear models,” Biometrika, 73(1): 13–22.
Moxnes, Erling and Eline van der Heijden (2003). “The effect of leadership in a public
bad experiment,” Journal of Conflict Resolution, 47(6): 773–795.
16
Potters, Jan, Martin Sefton, and Lise Vesterlund (2007). “Leading-by-example and signaling in voluntary contribution games: an experimental study,” Economic Theory,
33(1): 169–182.
Rivas, M. Fernanda and Matthias Sutter (2011). “The benefits of voluntary leadership in
experimental public goods games,” Economics Letters, 112(2): 176–178.
Salamon, Lester M., S. Wojciech Sokolowski, and Megan A. Haddock (2011). “Measuring the economic value of volunteer work globally: Concepts, estimates, and a
roadmap to the future,” Annals of Public and Cooperative Economics, 82(3): 217–
252.
17
8
12
Base
No compensation
10
Moderate compensation
1
2
3
4
5
6
Round
7
8
9
0
0
2
2
4
Average contribution
6
8
Average contribution ± s.e.
4
6
High compensation
10
Base
No comp Mod comp High comp
Figure 1: Contributions
Note: The left panel shows average contribution in each round by treatment. The
right panel shows average contribution over all rounds by treatment, where the
standard errors of the mean are also indicated (+/- 1 s.e.). Contributions of both
leaders and followers are included in both panels.
18
.8
Share ± s.e.
.4
.6
.2
0
0
.2
Share ± s.e.
.4
.6
.8
1
Share of groups with leader
1
Share of participants that volunteer
No comp
Mod comp
High comp
No comp
Mod comp
High comp
Figure 2: Recruitment
Note: The left panel shows the share of situations where a participant volunteered
to be a leader by treatment. The right panel shows the share of groups with a
leader by treatment. The standard errors of the mean are also indicated (+/- 1
s.e.).
19
12
10
Average contribution ± s.e.
4
6
8
2
0
No comp
Mod comp
High comp
Figure 3: Leader contribution
Note: The figure shows the average leader contribution by treatment. The standard
errors of the mean are also indicated (+/- 1 s.e.).
20
Share of freeriders that volunteer to be leader ± s.e.
.25
.5
.75
1
0
No comp
Mod comp
High comp
Figure 4: Leader composition across treatments
Note: The figure shows the share of free-riders (classified by the P-experiment)
that volunteered to be leader. The standard errors of the mean are also indicated
(+/- 1 s.e.).
21
Follower contribution
0 5 10 15 20
Moderate compensation
Follower contribution
0 5 10 15 20
High compensation
0
5
10
15
Leader contribution
20
0
20
Follower contribution
0 5 10 15 20
Total
Follower contribution
0 5 10 15 20
No compensation
5
10
15
Leader contribution
0
5
10
15
Leader contribution
20
0
5
10
15
Leader contribution
20
Figure 5: Follower contribution as a function of leader contribution
Note: The figure shows a scatter of follower contribution plotted against the
leader’s contribution by treatment and pooled. The solid line shows the linear
fit, the dashed line indicates the 45-degree line.
22
Strongly agree
Strongly disagree
Neutral
Mod compensation
High compensation
Inequality
Envy
Disappointment
Comp. to leader
Figure 6: Questionnaire
Note: The figure reports the average response from 1-5 indicating the extent to
which the participant agreed with different statements about the motivation for
the choices made in the experiment for moderate and high compensation. These
questions were only given in the 2013 fall sessions. The bars show the following:
Inequality: “For my contribution, considerations of inequality within the group
were very important for my decision of what to contribute.”, Envy: “When I attempted to become an early contributor and was not selected, I envied the leader.”,
Disappointment: “When I attempted to become an early contributor and was not
selected, I was very disappointed.” and Compensation to leader: “When there
was an early contributor, the fact that he was compensated for being an early contributor made me less motivated by his contribution.” The standard errors of the
mean are also indicated (+/- 1 s.e.).
23
.75
Share days volunteering ± s.e.
.55
.6
.65
.7
.5
No comp
Mod comp
High comp
Figure 7: Survey response: volunteer work
Note: The figure reports survey responses by treatment, where, for simplicity,
we refer to the treatment variation in the same way as in the lab experiments
(no/moderate/high compensation). In one round of the survey, the choice set was
zero to four days. We show the average share of days (out of four/six) that the
respondent said that he would have participated in voluntary work. The standard
errors of the mean are also indicated (+/- 1 s.e.).
24